Stay Informed
Sign up here for the latest articles
By Ravi Pau
Despite every dashboard saying AI is working, consumers are wishing brands hadn’t bothered. Ravi Pau, Havas’ Head of AI Operations, warns against The AI Efficiency Trap: confusing organisational efficiency with actual customer value.
Customer service cuts the time it takes to handle a query. Marketing doubles output. Ecommerce automates product recommendations. Operations cuts hours from the process. The numbers move in the right direction.
Then the customer meets it: the chatbot cannot solve the problem, the recommendation makes little sense. The business has become more efficient, but the experience has not become better.
Businesses are becoming very good at measuring what AI does for them – time saved, cost removed, output increased, processes automated. Far fewer are measuring what all that efficiency does to the person receiving it.
Havas surveyed 10,558 people across seven countries and evaluated more than 21,000 AI enabled brand experiences to understand that gap. Only 9.4% of the experiences reached the highest performance band.
Most AI experiences are not spectacular failures. They don’t go viral for the wrong reason, or cause board level crises. They are simply average: functional enough to approve, cheap enough to scale, but forgettable enough to do nothing for the brand.
Customers do not know that the recommendation comes from the ecommerce team, the chatbot from customer service or emails timed by CRM. They just experience one company.
AI is being deployed across organisations through dozens of individual business cases, each with its own owner and its own definition of success. Someone still has to ask what those decisions add up to when they reach the customer.
Those decisions are already showing up in how people judge brands. Half of people in our study said AI can make things look, feel, or sound the same. More than half (56%) said its use by brands can feel misleading. Less than third (29%) agreed that brands’ use of AI feels human.
I work in Operations. If AI removes 20 hours from the process, I want those hours. The customer, however, only cares when something becomes easier for them. Did they understand the product? Was it easier to get an AI generated product recommendation? Did the problem get solved?
When a company gets the efficiency and the customer gets little in return, we call it the AI Efficiency Trap. It is not an argument against efficiency. It is an argument against confusing efficiency with value. They are not the same scorecard. That matters even more because there is a bigger brand problem here too.
Havas’ Science of Desire research found that 84% of brands suffer from indifference. They are known but not strongly wanted; familiar, but replaceable. At the same time, half of people in our AI study said that the technology can make things look, feel, or sound the same.
Put those together and AI starts to look less like an automatic answer to the brand problem. Every company is gaining access to similar models, producing more content, recommendations and interactions for less money.
If those systems optimise towards the same kind of outputs, the risk is clear: AI does not just make good work cheaper, it makes average work easier to scale.
For a CMO, that should be more worrying than a failed experiment. Failure gets noticed and fixed, while mediocrity sits comfortably inside a dashboard full of green numbers. A content team could double its output and call it a success, but the harder question is whether anyone wanted twice of much as what it was making.
We created the Valuable AI Experience Score to move beyond adoption and efficiency metrics and start measuring what actually matters to the customer: whether the experience is useful, credible and worth their attention. Productivity measures what AI saved the company. The Score measures what the customer got back.
High scoring experiences in the study were associated with 4.4x higher Engagement, 1.6x stronger Brand Desire, and 1.5x stronger Growth scores than low scoring experiences.
The pattern became clearest when we looked at what AI was actually being asked to do. It performed best when helping people understand, compare and get support. Creator style AI content came last.
Automotive was the strongest sector we tested. Healthcare over the counter was the lowest. This doesn’t mean people love AI when buying cars and reject it when buying a headache tablet. But it means that the job, the stakes, and the expectations change.
Buying a car involves models, features, prices and trade offs. There is plenty for AI to organise, compare and explain. Healthcare carries a different burden around accuracy and responsibility. The customer applies a higher standard because getting it wrong means something different.
The obvious objection to all of this is that customers do not care whether AI was involved if the experiences works. Our qualitative research found that people often did not even notice AI when the output was coherent and useful.
AI does not need to announce itself. The customer shouldn’t leave thinking, ‘What an excellent use of AI.’ They should leave thinking that the problem was solved quickly, the explanation made sense or the choice became easier.
Nobody celebrates the cloud infrastructure behind a good checkout experience. AI should be treated with the same lack of ceremony. That is why customer facing AI cannot be governed entirely as a series of technology projects.
A business can have 30 successful AI deployments, all achieving their local targets, that collectively make the brand more generic, colder, or harder to deal with. If each project is judged only against its own cost saving or productivity target, nobody is responsible for the total effect.
The first phase of AI was about getting access to the technology and proving that it could make business faster. That argument has been (largely) won.
The next phase is harder because access does not differentiate anyone. Most serious companies will have similar models, similar automation and similar ability to generate things at scale.
The advantage will come from being more selective. Knowing where AI genuinely improves an experience, where to turn the dial between human and machine and where adding AI would solve a problem that nobody actually had.
95% of businesses have AI strategies but only 9.4% of experiences that we tested are reaching the highest band. This is telling us that every dashboard is saying that AI is working, but customers are wishing you hadn’t bothered.
About the author
Ravi Pau is Head of AI Operations at Havas
Havas’ full Valuable AI Experience research, including sector benchmarks and market data is available here.